Stuart Russell is a computer scientist, a Distinguished Professor of Computer Science at the University of California, Berkeley, and the co-author, with Peter Norvig, of the textbook Artificial Intelligence: A Modern Approach. He directs Berkeley's Center for Human-Compatible AI and wrote Human Compatible: AI and the Problem of Control (Viking/Penguin, 2019). The label "father of modern artificial intelligence" does not appear in the Berkeley biography or the Royal Society profile checked for this article, so it is better read as praise than as a documented title.
Key takeaways
- Russell earned a B.A. in physics at Oxford in 1982 and a Ph.D. in computer science at Stanford in 1986, then joined the Berkeley faculty.
- His textbook with Peter Norvig is now in a fourth U.S. edition; his Berkeley biography says it has been used at 1,500 universities in 135 countries.
- He gave the 2021 BBC Reith Lectures, was appointed an OBE in 2021, and is listed by the Royal Society as a Fellow elected in 2025.
- His research proposal is that machines should be uncertain about human preferences and learn them from behavior, rather than optimize a fixed objective.
- That proposal is a research direction, and the evidence cited here is theoretical papers, not a deployed standard.
Career in brief
According to his official biography, Russell holds a first-class honours B.A. in physics from Oxford (1982) and a Ph.D. in computer science from Stanford (1986). He is a Distinguished Professor of Electrical Engineering and Computer Sciences at Berkeley, holds the Smith-Zadeh Chair in Engineering, directs the Center for Human-Compatible AI and the Kavli Center for Ethics, Science, and the Public, and is also a professor of computational precision health at the University of California, San Francisco.
The same page lists memberships including the U.S. National Academy of Engineering and fellowships of AAAI, ACM, and AAAS, plus awards such as the IJCAI Computers and Thought Award and the ACM Allen Newell Award. His listed research areas include machine learning, probabilistic reasoning, knowledge representation, planning, and the philosophical foundations of AI. The Royal Society lists him as Professor Stuart Russell OBE FRS, with election in 2025.
The textbook
Artificial Intelligence: A Modern Approach, written with Peter Norvig, is the work most students meet first. The book's Berkeley site shows a fourth U.S. edition, and the table of contents includes a chapter titled Search in Complex Environments and another on constraint satisfaction problems. Russell's biography states that the book has been translated into 14 languages and is used at 1,500 universities in 135 countries. The book's site separately says it has been adopted by over 1,500 schools. Adoption counts are self-reported and were not independently audited for this article.
The control problem and the Reith Lectures
Human Compatible, published in 2019, sets out Russell's argument about control. In 2021 he gave the BBC Reith Lectures under the title Living With Artificial Intelligence, in four parts: The Biggest Event in Human History, AI in Warfare, AI in the Economy, and AI: A Future for Humans. His biography lists autonomous weapons and the long-term future of AI among his current concerns.
The Center for Human-Compatible AI states its aim as shifting AI research toward provably beneficial systems and away from machines that simply maximize arbitrary objectives. The claim in the book and the lectures is not a prediction that a specific system will fail on a specific date. It is an argument about how objectives should be specified.
Two technical ideas, with a worked illustration
The first idea is inverse reinforcement learning. In a 2000 paper presented at ICML (pages 663 to 670), Andrew Ng and Russell addressed the problem of extracting a reward function from observed, optimal behavior, deriving three algorithms for different cases. Instead of being told what to want, the system infers it from what an agent does.
The second is the off-switch game, an arXiv paper first posted on November 24, 2016 by Dylan Hadfield-Menell, Anca Dragan, Pieter Abbeel, and Russell. It analyzes whether a robot has a reason to allow a human to switch it off. The authors find that an agent that treats its reward function as given has an incentive to disable the switch, except when the human is perfectly rational, and that a robot uncertain about the utility of outcomes, and treating human actions as evidence, can be safer.
An illustrative calculation, not taken from the paper, shows the logic. Suppose a robot is equally unsure whether a proposed action is worth +1 or -1 to its owner. Acting at once has an expected value of 0. If the robot instead defers, a rational owner blocks the action when it is worth -1 and allows it when it is worth +1, so the expected value is 0.5. Uncertainty makes deferring worth more than acting alone. If the robot were certain the action was worth +1, deferring would add nothing, which is the intuition behind the paper's contrast with fixed-objective agents.
| Idea | Source | What it claims | Limit |
|---|---|---|---|
| Inverse reinforcement learning | Ng and Russell, ICML 2000 | A reward function can be inferred from observed optimal behavior | Assumes the observed behavior is optimal |
| Off-switch game | Hadfield-Menell, Dragan, Abbeel, Russell, arXiv 2016 | An agent uncertain about its objective has a reason to allow shutdown | Analysis of a simple game, not a deployed system |
FAQ
Is Stuart Russell the father of AI?
No source checked gives him that title. He is better described as a leading textbook author and researcher on how AI systems should be given objectives.
What does the Center for Human-Compatible AI do?
It is based at UC Berkeley and says its mission is to move AI research toward provably beneficial systems. Its page does not list a founding date, so none is given here.
What is Human Compatible about?
It is a 2019 book on the problem of keeping increasingly capable AI systems under human control, built around the idea that machines should be uncertain about human preferences.
Does his approach exist in products today?
The sources checked present it as research. No deployed commercial system was verified as implementing it, so no such claim is made.
Related reading
- Who gets called the father of AI, and why
- Marvin Minsky and early cognitive science
- The debate over AI as a threat
- Traditional AI and machine learning compared
More in AI Basics.
Last verified: October 2026
Sources
- Stuart Russell, Biography, University of California, Berkeley (official faculty page)
- Professor Stuart Russell OBE FRS, The Royal Society, fellow profile
- 2021 Reith Lectures: Living With Artificial Intelligence, Stuart Russell, BBC, listing on his Berkeley page
- Algorithms for Inverse Reinforcement Learning, Ng A, Russell S, ICML 2000, pp. 663-670
- The Off-Switch Game, Hadfield-Menell D, Dragan A, Abbeel P, Russell S, arXiv:1611.08219, 2016
- Artificial Intelligence: A Modern Approach, Russell S, Norvig P, book website, UC Berkeley; Center for Human-Compatible AI, About page